Many animals and humans process the visual field with a varying spatial resolution (foveated vision) and use peripheral processing to make eye movements and point the fovea to acquire high-resolution information about objects of interest. This architecture results in computationally efficient rapid scene exploration. Recent progress in vision Transformers has brought about new alternatives to the traditionally convolution-reliant computer vision systems. However, these models do not explicitly model the foveated properties of the visual system nor the interaction between eye movements and the classification task. We propose foveated Transformer (FoveaTer) model, which uses pooling regions and saccadic movements to perform object classification tasks using a vision Transformer architecture. Our proposed model pools the image features using squared pooling regions, an approximation to the biologically-inspired foveated architecture, and uses the pooled features as an input to a Transformer Network. It decides on the following fixation location based on the attention assigned by the Transformer to various locations from previous and present fixations. The model uses a confidence threshold to stop scene exploration, allowing to dynamically allocate more fixation/computational resources to more challenging images. We construct an ensemble model using our proposed model and unfoveated model, achieving an accuracy 1.36% below the unfoveated model with 22% computational savings. Finally, we demonstrate our model's robustness against adversarial attacks, where it outperforms the unfoveated model.
In open question answering (QA), the answer to a question is produced by retrieving and then analyzing documents that might contain answers to the question. Most open QA systems have considered only retrieving information from unstructured text. Here we consider for the first time open QA over both tabular and textual data and present a new large-scale dataset Open Table-Text Question Answering (OTT-QA) to evaluate performance on this task. Most questions in OTT-QA require multi-hop inference across tabular data and unstructured text, and the evidence required to answer a question can be distributed in different ways over these two types of input, making evidence retrieval challenging---our baseline model using an iterative retriever and BERT-based reader achieves an exact match score less than 10%. We then propose two novel techniques to address the challenge of retrieving and aggregating evidence for OTT-QA. The first technique is to use "early fusion" to group multiple highly relevant tabular and textual units into a fused block, which provides more context for the retriever to search for. The second technique is to use a cross-block reader to model the cross-dependency between multiple retrieved evidences with global-local sparse attention. Combining these two techniques improves the score significantly, to above 27%.
Generative adversarial networks (GANs) learn the distribution of observed samples through a zero-sum game between two machine players, a generator and a discriminator. While GANs achieve great success in learning the complex distribution of image, sound, and text data, they perform suboptimally in learning multi-modal distribution-learning benchmarks including Gaussian mixture models (GMMs). In this paper, we propose Generative Adversarial Training for Gaussian Mixture Models (GAT-GMM), a minimax GAN framework for learning GMMs. Motivated by optimal transport theory, we design the zero-sum game in GAT-GMM using a random linear generator and a softmax-based quadratic discriminator architecture, which leads to a non-convex concave minimax optimization problem. We show that a Gradient Descent Ascent (GDA) method converges to an approximate stationary minimax point of the GAT-GMM optimization problem. In the benchmark case of a mixture of two symmetric, well-separated Gaussians, we further show this stationary point recovers the true parameters of the underlying GMM. We numerically support our theoretical findings by performing several experiments, which demonstrate that GAT-GMM can perform as well as the expectation-maximization algorithm in learning mixtures of two Gaussians.
Existing question answering datasets focus on dealing with homogeneous information, based either only on text or KB/Table information alone. However, as human knowledge is distributed over heterogeneous forms, using homogeneous information might lead to severe coverage problems. To fill in the gap, we present \dataset, a new large-scale question-answering dataset that requires reasoning on heterogeneous information. Each question is aligned with a structured Wikipedia table and multiple free-form corpora linked with the entities in the table. The questions are designed to aggregate both tabular information and text information, i.e. lack of either form would render the question unanswerable. We test with three different models: 1) table-only model. 2) text-only model. 3) a hybrid model \model which combines both table and textual information to build a reasoning path towards the answer. The experimental results show that the first two baselines obtain compromised scores below 20\%, while \model significantly boosts EM score to over 50\%, which proves the necessity to aggregate both structure and unstructured information in \dataset. However, \model's score is still far behind human performance, hence we believe \dataset to an ideal and challenging benchmark to study question answering under heterogeneous information. The dataset and code are available at \url{https://github.com/wenhuchen/HybridQA}.
There are two main lines of research on visual reasoning: neural module network (NMN) with explicit multi-hop reasoning through handcrafted neural modules, and monolithic network with implicit reasoning in the latent feature space. The former excels in interpretability and compositionality, while the latter usually achieves better performance due to model flexibility and parameter efficiency. In order to bridge the gap between the two and leverage the merits of both, we present Meta Module Network (MMN), a novel hybrid approach that can utilize a Meta Module to perform versatile functionalities, while preserving compositionality and interpretability through modularized design. The proposed model first parses an input question into a functional program through a Program Generator. Instead of handcrafting a task-specific network to represent each function similar to traditional NMN, we propose a Meta Module, which can read a recipe (function specifications) to dynamically instantiate the task-specific Instance Modules for compositional reasoning. To endow different instance modules with designated functionalities, we design a symbolic teacher which can execute against provided scene graphs to generate guidelines for the instantiated modules (student) to follow during training. Experiments conducted on the GQA benchmark demonstrates that MMN outperforms both NMN and monolithic network baselines, with good generalization ability to handle unseen functions.
Modelling relations between multiple entities has attracted increasing attention recently, and a new dataset called DocRED has been collected in order to accelerate the research on the document-level relation extraction. Current baselines for this task uses BiLSTM to encode the whole document and are trained from scratch. We argue that such simple baselines are not strong enough to model to complex interaction between entities. In this paper, we further apply a pre-trained language model (BERT) to provide a stronger baseline for this task. We also find that solving this task in phases can further improve the performance. The first step is to predict whether or not two entities have a relation, the second step is to predict the specific relation.
With the rapid development in deep learning, deep neural networks have been widely adopted in many real-life natural language applications. Under deep neural networks, a pre-defined vocabulary is required to vectorize text inputs. The canonical approach to select pre-defined vocabulary is based on the word frequency, where a threshold is selected to cut off the long tail distribution. However, we observed that such simple approach could easily lead to under-sized vocabulary or over-sized vocabulary issues. Therefore, we are interested in understanding how the end-task classification accuracy is related to the vocabulary size and what is the minimum required vocabulary size to achieve a specific performance. In this paper, we provide a more sophisticated variational vocabulary dropout (VVD) based on variational dropout to perform vocabulary selection, which can intelligently select the subset of the vocabulary to achieve the required performance. To evaluate different algorithms on the newly proposed vocabulary selection problem, we propose two new metrics: Area Under Accuracy-Vocab Curve and Vocab Size under X\% Accuracy Drop. Through extensive experiments on various NLP classification tasks, our variational framework is shown to significantly outperform the frequency-based and other selection baselines on these metrics.
Despite the great success object detection and segmentation models have achieved in recognizing individual objects in images, performance on cognitive tasks such as image caption, semantic image retrieval, and visual QA is far from satisfactory. To achieve better performance on these cognitive tasks, merely recognizing individual object instances is insufficient. Instead, the interactions between object instances need to be captured in order to facilitate reasoning and understanding of the visual scenes in an image. Scene graph, a graph representation of images that captures object instances and their relationships, offers a comprehensive understanding of an image. However, existing techniques on scene graph generation fail to distinguish subjects and objects in the visual scenes of images and thus do not perform well with real-world datasets where exist ambiguous object instances. In this work, we propose a novel scene graph generation model for predicting object instances and its corresponding relationships in an image. Our model, SG-CRF, learns the sequential order of subject and object in a relationship triplet, and the semantic compatibility of object instance nodes and relationship nodes in a scene graph efficiently. Experiments empirically show that SG-CRF outperforms the state-of-the-art methods, on three different datasets, i.e., CLEVR, VRD, and Visual Genome, raising the Recall@100 from 24.99% to 49.95%, from 41.92% to 50.47%, and from 54.69% to 54.77%, respectively.
With the recent surge of interests in deep neural networks, more real-world applications start to adopt it in practice. However, deep neural networks are known to have limited control over its prediction under unseen images. Such weakness can potentially threaten society and cause annoying consequences in real-world scenarios. In order to resolve such issue, a popular task called out-of-distribution detection was proposed, which aims at separating out-of-distribution images from in-distribution images. In this paper, we propose a perturbed prior network architecture, which can efficiently separate model-level uncertainty from data-level uncertainty via prior entropy. To further enhance the robustness of proposed entropy-based uncertainty measure, we propose a concentration perturbation algorithm, which adaptively adds noise to concentration parameters so that the in- and out-of-distribution images are better separable. Our method can directly rely on the pre-trained deep neural network without re-training it, and also requires no knowledge about the network architecture and out-of-distribution examples. Such simplicity makes our method more suitable for real-world AI applications. Through comprehensive experiments, our methods demonstrate its superiority by achieving state-of-the-art results on many datasets.